New model releases target video, cybersecurity and voice agents

October 2026 model launches span coding, computer use, video generation, robotics and enterprise speech. Pika announced H3 Flex on Oct 29, a post-trained version of MiniMax H3 for iterative video creation. Pika says it supports up to 12 multimodal references, synchronized audio and open weights for local use. OpenAI released GPT-6.1 Terra on Oct 26, combining visual reasoning and computer use with clearer writing. OpenAI says standard API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.

Google DeepMind introduced Gemini 4 Carbon for cybersecurity defense and long-horizon tasks, initially limiting access to selected defenders through its Fairwind Program. Broader availability depends on further guardrail work. Its Genie 4 world model supports real-time interactive simulation with native audio and natural-language scene control, but remains limited to trusted testers. Gemini Robotics ER 2 is available in public preview through Google Cloud.

Other launches address specialized deployment needs. Rime’s Prism provides enterprise speech-to-speech interaction with inspectable text before speech and tool calls alongside spoken responses. Particle-1 targets software engineering, while Liquid AI’s open-weight LFM3 family shares a backbone across text, vision, audio and decision models. Availability varies across previews, hosted APIs and downloadable weights, making task-specific checks of quality, latency, cost and human-review requirements important before adoption.